AIR#86 - May 29, 2024
Good morning, AI enthusiasts! Grab your coffee and get ready to dive into today's edition of AIR: The AI Recon, where the world of artificial intelligence is buzzing with stories that will both intrigue and challenge you. Leading the headlines is the revelation that Transformers can perform arithmetic with high accuracy using position-aware embeddings. This breakthrough could significantly enhance the performance of AI on complex tasks, making it a must-read for anyone interested in the cutting-edge of AI research.
But it's not all about technical marvels. Facebook is making waves with its controversial decision to use your photos and posts to train its AI, sparking major privacy concerns. The opt-out process is anything but straightforward, leaving users in a tricky spot. Meanwhile, in a dramatic twist, leaked documents reveal that Sam Altman knew about silencing former OpenAI employees with non-disclosure threats, despite his public denials. This scandal has cast a shadow over Altman's leadership and raises serious questions about transparency and ethics in AI firms.
As we explore these stories, let's not forget the broader implications they bring to the table. From Jan Leike joining Anthropic's superalignment team to advance scalable oversight, to the Chatbot Arena Leaderboard where Gemini 1.5 is challenging GPT-4, today's edition is packed with insights that challenge our understanding of AI's role in society. Whether it's the promise of new technologies or the ethical dilemmas they pose, there's plenty to ponder as we delve into the dynamic world of artificial intelligence together. So, sit back, sip your coffee, and let's explore these captivating narratives.
Business
Facebook to Use Your Photos and Posts to Train AI (Opt-Out is Complicated)
Facebook will use your posts and photos to train its AI, but opting out is a complex process. Privacy concerns are rising.
🔥 Leaked Docs Reveal Sam Altman Knew About Silencing Former OpenAI Employees
Leaked docs show Sam Altman knew about silencing ex-OpenAI employees with non-disclosure threats, despite his public denials.
🔥 Ex-OpenAI Board Member Reveals Sam Altman's Lies, Leading to Brief Ousting
Ex-OpenAI board member Helen Toner claims Sam Altman lied to the board, leading to his brief ousting as CEO.
Get AI-Ready Data in Minutes with Tonic Textual
Get AI-ready data in minutes with Tonic Textual: extract, govern, enrich, and deploy unstructured data securely for LLMs.
Engineering
🔥 [Paper] Transformers Can Do Arithmetic with the Right Embeddings
Transformers can perform arithmetic with high accuracy by using position-aware embeddings, enabling better performance on complex tasks.
Jan Leike Joins Anthropic's Superalignment Team
Jan Leike joins Anthropic's superalignment team to advance scalable oversight and automated alignment research. Interested? DM him!
Chatbot Arena Leaderboard: Gemini 1.5 Challenges GPT-4
Gemini 1.5 Pro/Advanced ranks #2, challenging GPT-4, while Flash hits #9, outperforming Llama-3-70b. Major leap for AI!
🔥 [GitHub] Tinygrad 0.9.0 Released: Major Updates & New Features
Tinygrad 0.9.0 released with new docs, GPU backends, Nvidia tensor core support, and faster training. Check the full changelog on GitHub!
🔥 [GitHub] Reproducing GPT-2 in C/CUDA in 90 Minutes for $20 by Karpathy
Karpathy reproduces GPT-2 (124M) in C/CUDA in 90 minutes for just $20, making advanced AI training more accessible. 🚀
🔥 Llama 3-V: GPT-4V Performance for $500 with 100x Smaller Model
Llama 3-V matches GPT-4V's performance with a model 100x smaller, trained for just $500, offering a 10-20% performance boost over Llava.